I´d assume that IMPORT should work in parallel automatically:
What surprised me was that IMPORT did perform slower than CTAS on a 50M copy...
|COMMIT||1.208||0||COMMIT /* AUTO */|
|CREATE TABLE AS||1312.431||50000000||create table t1 as select level as my_pk,HASH_MD5(to_char(level)) as some_text,current_date + mod(level,100000) as some_date,'fix text for good measure' as some_fix_text from dual connect by level<=50000000;|
fyi: not the most ideal/performant why to create the data but I wanted to use a sql that can by copy&pasted&executed
Copy with CTAS:
|COMMIT||1.845||0||COMMIT /* AUTO */|
|CREATE TABLE AS||1.013||50000000||create table t2 as select * From t1;|
Copy with IMPORT:
|COMMIT||1.318||0||COMMIT /* AUTO */|
|IMPORT||5.256||50000000||import into t3 from exa at ... table t1;|
I´d assume that on a table with a bigger size and more columns IMPORT should outperform CTAS ( which btw also works as parallel as your cluster can muster i.e. #nodes) ..
Depending on what "for experimenting" means in your case you might save a bit if you can manage to stay in the same session for copy&experimenting and not commit your copy ( i.e. CTAS or IMPORT with autocommit off and all tests in the same transaction ) - of course, that will take a toll on your DBRAM and is not viable if your tests can´t be made in the same session.
> I´d assume that IMPORT should work in parallel automatically.
If you import from EXA as data source type, yes.
> What surprised me was that IMPORT did perform slower than CTAS on a 50M copy...
Of course CTAS/INSERT also work in parallel and can do without any driver data conversion overhead...
The only situation where IMPORT used to be (?) faster than INSERT was when the target table had a distribution key differing from the source table; with IMPORT the distribution was done in-stream, while the INSERT performed a two-phase approach.
All in all, CTAS should be the fastest possible way to copy a data set. If the data set does not fully fit into memory, maybe a loop of smaller INSERT statements is preferable.